brainstate 0.1.10__py2.py3-none-any.whl → 0.2.1__py2.py3-none-any.whl

brainstate/nn/_dropout.py

@@ -1,426 +1,618 @@
-# Copyright 2024 BDP Ecosystem Limited. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-# ==============================================================================
-
-
-from functools import partial
-from typing import Optional, Sequence
-
-import brainunit as u
-import jax.numpy as jnp
-
-from brainstate import random, environ, init
-from brainstate._state import ShortTermState
-from brainstate.typing import Size
-from ._module import ElementWiseBlock
-
-__all__ = [
-    'DropoutFixed', 'Dropout', 'Dropout1d', 'Dropout2d', 'Dropout3d',
-]
-
-
-class Dropout(ElementWiseBlock):
-    """A layer that stochastically ignores a subset of inputs each training step.
-
-    In training, to compensate for the fraction of input values dropped (`rate`),
-    all surviving values are multiplied by `1 / (1 - rate)`.
-
-    This layer is active only during training (``mode=brainstate.mixin.Training``). In other
-    circumstances it is a no-op.
-
-    .. [1] Srivastava, Nitish, et al. "Dropout: a simple way to prevent
-           neural networks from overfitting." The journal of machine learning
-           research 15.1 (2014): 1929-1958.
-
-    Args:
-        prob: Probability to keep element of the tensor.
-        broadcast_dims: dimensions that will share the same dropout mask.
-        name: str. The name of the dynamic system.
-
-    """
-    __module__ = 'brainstate.nn'
-
-    def __init__(
-        self,
-        prob: float = 0.5,
-        broadcast_dims: Sequence[int] = (),
-        name: Optional[str] = None
-    ) -> None:
-        super().__init__(name=name)
-        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
-        self.prob = prob
-        self.broadcast_dims = broadcast_dims
-
-    def __call__(self, x):
-        dtype = u.math.get_dtype(x)
-        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
-        if fit_phase and self.prob < 1.:
-            broadcast_shape = list(x.shape)
-            for dim in self.broadcast_dims:
-                broadcast_shape[dim] = 1
-            keep_mask = random.bernoulli(self.prob, broadcast_shape)
-            keep_mask = u.math.broadcast_to(keep_mask, x.shape)
-            return u.math.where(
-                keep_mask,
-                u.math.asarray(x / self.prob, dtype=dtype),
-                u.math.asarray(0., dtype=dtype)
-            )
-        else:
-            return x
-
-
-class _DropoutNd(ElementWiseBlock):
-    __module__ = 'brainstate.nn'
-    prob: float
-    channel_axis: int
-    minimal_dim: int
-
-    def __init__(
-        self,
-        prob: float = 0.5,
-        channel_axis: int = -1,
-        name: Optional[str] = None
-    ) -> None:
-        super().__init__(name=name)
-        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
-        self.prob = prob
-        self.channel_axis = channel_axis
-
-    def __call__(self, x):
-        # check input shape
-        inp_dim = u.math.ndim(x)
-        if inp_dim not in (self.minimal_dim, self.minimal_dim + 1):
-            raise RuntimeError(f"dropout1d: Expected {self.minimal_dim}D or {self.minimal_dim + 1}D input, "
-                               f"but received a {inp_dim}D input. {self._get_msg(x)}")
-        is_not_batched = self.minimal_dim
-        if is_not_batched:
-            channel_axis = self.channel_axis if self.channel_axis >= 0 else (x.ndim + self.channel_axis)
-            mask_shape = [(dim if i == channel_axis else 1) for i, dim in enumerate(x.shape)]
-        else:
-            channel_axis = (self.channel_axis + 1) if self.channel_axis >= 0 else (x.ndim + self.channel_axis)
-            assert channel_axis != 0, f"Channel axis must not be 0. But got {self.channel_axis}."
-            mask_shape = [(dim if i in (channel_axis, 0) else 1) for i, dim in enumerate(x.shape)]
-
-        # get fit phase
-        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
-
-        # generate mask
-        if fit_phase and self.prob < 1.:
-            dtype = u.math.get_dtype(x)
-            keep_mask = random.bernoulli(self.prob, mask_shape)
-            keep_mask = jnp.broadcast_to(keep_mask, x.shape)
-            return jnp.where(
-                keep_mask,
-                jnp.asarray(x / self.prob, dtype=dtype),
-                jnp.asarray(0., dtype=dtype)
-            )
-        else:
-            return x
-
-    def _get_msg(self, x):
-        return ''
-
-
-class Dropout1d(_DropoutNd):
-    r"""Randomly zero out entire channels (a channel is a 1D feature map,
-    e.g., the :math:`j`-th channel of the :math:`i`-th sample in the
-    batched input is a 1D tensor :math:`\text{input}[i, j]`).
-    Each channel will be zeroed out independently on every forward call with
-    probability :attr:`p` using samples from a Bernoulli distribution.
-
-    Usually the input comes from :class:`nn.Conv1d` modules.
-
-    As described in the paper
-    `Efficient Object Localization Using Convolutional Networks`_ ,
-    if adjacent pixels within feature maps are strongly correlated
-    (as is normally the case in early convolution layers) then i.i.d. dropout
-    will not regularize the activations and will otherwise just result
-    in an effective learning rate decrease.
-
-    In this case, :func:`nn.Dropout1d` will help promote independence between
-    feature maps and should be used instead.
-
-    Args:
-        prob: float. probability of an element to be zero-ed.
-
-    Shape:
-        - Input: :math:`(N, C, L)` or :math:`(C, L)`.
-        - Output: :math:`(N, C, L)` or :math:`(C, L)` (same shape as input).
-
-    Examples::
-
-        >>> m = Dropout1d(p=0.2)
-        >>> x = random.randn(20, 32, 16)
-        >>> output = m(x)
-        >>> output.shape
-        (20, 32, 16)
-
-    .. _Efficient Object Localization Using Convolutional Networks:
-       https://arxiv.org/abs/1411.4280
-    """
-    __module__ = 'brainstate.nn'
-    minimal_dim: int = 2
-
-    def _get_msg(self, x):
-        return ("Note that dropout1d exists to provide channel-wise dropout on inputs with 1 "
-                "spatial dimension, a channel dimension, and an optional batch dimension "
-                "(i.e. 2D or 3D inputs).")
-
-
-class Dropout2d(_DropoutNd):
-    r"""Randomly zero out entire channels (a channel is a 2D feature map,
-    e.g., the :math:`j`-th channel of the :math:`i`-th sample in the
-    batched input is a 2D tensor :math:`\text{input}[i, j]`).
-    Each channel will be zeroed out independently on every forward call with
-    probability :attr:`p` using samples from a Bernoulli distribution.
-
-    Usually the input comes from :class:`nn.Conv2d` modules.
-
-    As described in the paper
-    `Efficient Object Localization Using Convolutional Networks`_ ,
-    if adjacent pixels within feature maps are strongly correlated
-    (as is normally the case in early convolution layers) then i.i.d. dropout
-    will not regularize the activations and will otherwise just result
-    in an effective learning rate decrease.
-
-    In this case, :func:`nn.Dropout2d` will help promote independence between
-    feature maps and should be used instead.
-
-    Args:
-        prob: float. probability of an element to be kept.
-
-    Shape:
-        - Input: :math:`(N, C, H, W)` or :math:`(N, C, L)`.
-        - Output: :math:`(N, C, H, W)` or :math:`(N, C, L)` (same shape as input).
-
-    Examples::
-
-        >>> m = Dropout2d(p=0.2)
-        >>> x = random.randn(20, 32, 32, 16)
-        >>> output = m(x)
-
-    .. _Efficient Object Localization Using Convolutional Networks:
-       https://arxiv.org/abs/1411.4280
-    """
-    __module__ = 'brainstate.nn'
-    minimal_dim: int = 3
-
-    def _get_msg(self, x):
-        return ("Note that dropout2d exists to provide channel-wise dropout on inputs with 2 "
-                "spatial dimensions, a channel dimension, and an optional batch dimension "
-                "(i.e. 3D or 4D inputs).")
-
-
-class Dropout3d(_DropoutNd):
-    r"""Randomly zero out entire channels (a channel is a 3D feature map,
-    e.g., the :math:`j`-th channel of the :math:`i`-th sample in the
-    batched input is a 3D tensor :math:`\text{input}[i, j]`).
-    Each channel will be zeroed out independently on every forward call with
-    probability :attr:`p` using samples from a Bernoulli distribution.
-
-    Usually the input comes from :class:`nn.Conv3d` modules.
-
-    As described in the paper
-    `Efficient Object Localization Using Convolutional Networks`_ ,
-    if adjacent pixels within feature maps are strongly correlated
-    (as is normally the case in early convolution layers) then i.i.d. dropout
-    will not regularize the activations and will otherwise just result
-    in an effective learning rate decrease.
-
-    In this case, :func:`nn.Dropout3d` will help promote independence between
-    feature maps and should be used instead.
-
-    Args:
-        prob: float. probability of an element to be kept.
-
-    Shape:
-        - Input: :math:`(N, C, D, H, W)` or :math:`(C, D, H, W)`.
-        - Output: :math:`(N, C, D, H, W)` or :math:`(C, D, H, W)` (same shape as input).
-
-    Examples::
-
-        >>> m = Dropout3d(p=0.2)
-        >>> x = random.randn(20, 16, 4, 32, 32)
-        >>> output = m(x)
-
-    .. _Efficient Object Localization Using Convolutional Networks:
-       https://arxiv.org/abs/1411.4280
-    """
-    __module__ = 'brainstate.nn'
-    minimal_dim: int = 4
-
-    def _get_msg(self, x):
-        return ("Note that dropout3d exists to provide channel-wise dropout on inputs with 3 "
-                "spatial dimensions, a channel dimension, and an optional batch dimension "
-                "(i.e. 4D or 5D inputs).")
-
-
-class AlphaDropout(_DropoutNd):
-    r"""Applies Alpha Dropout over the input.
-
-    Alpha Dropout is a type of Dropout that maintains the self-normalizing
-    property.
-    For an input with zero mean and unit standard deviation, the output of
-    Alpha Dropout maintains the original mean and standard deviation of the
-    input.
-    Alpha Dropout goes hand-in-hand with SELU activation function, which ensures
-    that the outputs have zero mean and unit standard deviation.
-
-    During training, it randomly masks some of the elements of the input
-    tensor with probability *p* using samples from a bernoulli distribution.
-    The elements to masked are randomized on every forward call, and scaled
-    and shifted to maintain zero mean and unit standard deviation.
-
-    During evaluation the module simply computes an identity function.
-
-    More details can be found in the paper `Self-Normalizing Neural Networks`_ .
-
-    Args:
-        prob: float. probability of an element to be kept.
-
-    Shape:
-        - Input: :math:`(*)`. Input can be of any shape
-        - Output: :math:`(*)`. Output is of the same shape as input
-
-    Examples::
-
-        >>> m = AlphaDropout(p=0.2)
-        >>> x = random.randn(20, 16)
-        >>> output = m(x)
-
-    .. _Self-Normalizing Neural Networks: https://arxiv.org/abs/1706.02515
-    """
-    __module__ = 'brainstate.nn'
-
-    def update(self, *args, **kwargs):
-        raise NotImplementedError("AlphaDropout is not supported in the current version.")
-
-
-class FeatureAlphaDropout(_DropoutNd):
-    r"""Randomly masks out entire channels (a channel is a feature map,
-    e.g. the :math:`j`-th channel of the :math:`i`-th sample in the batch input
-    is a tensor :math:`\text{input}[i, j]`) of the input tensor). Instead of
-    setting activations to zero, as in regular Dropout, the activations are set
-    to the negative saturation value of the SELU activation function. More details
-    can be found in the paper `Self-Normalizing Neural Networks`_ .
-
-    Each element will be masked independently for each sample on every forward
-    call with probability :attr:`p` using samples from a Bernoulli distribution.
-    The elements to be masked are randomized on every forward call, and scaled
-    and shifted to maintain zero mean and unit variance.
-
-    Usually the input comes from :class:`nn.AlphaDropout` modules.
-
-    As described in the paper
-    `Efficient Object Localization Using Convolutional Networks`_ ,
-    if adjacent pixels within feature maps are strongly correlated
-    (as is normally the case in early convolution layers) then i.i.d. dropout
-    will not regularize the activations and will otherwise just result
-    in an effective learning rate decrease.
-
-    In this case, :func:`nn.AlphaDropout` will help promote independence between
-    feature maps and should be used instead.
-
-    Args:
-        prob: float. probability of an element to be kept.
-
-    Shape:
-        - Input: :math:`(N, C, D, H, W)` or :math:`(C, D, H, W)`.
-        - Output: :math:`(N, C, D, H, W)` or :math:`(C, D, H, W)` (same shape as input).
-
-    Examples::
-
-        >>> m = FeatureAlphaDropout(p=0.2)
-        >>> x = random.randn(20, 16, 4, 32, 32)
-        >>> output = m(x)
-
-    .. _Self-Normalizing Neural Networks: https://arxiv.org/abs/1706.02515
-    .. _Efficient Object Localization Using Convolutional Networks:
-       https://arxiv.org/abs/1411.4280
-    """
-    __module__ = 'brainstate.nn'
-
-    def update(self, *args, **kwargs):
-        raise NotImplementedError("FeatureAlphaDropout is not supported in the current version.")
-
-
-class DropoutFixed(ElementWiseBlock):
-    """
-    A dropout layer with the fixed dropout mask along the time axis once after initialized.
-
-    In training, to compensate for the fraction of input values dropped (`rate`),
-    all surviving values are multiplied by `1 / (1 - rate)`.
-
-    This layer is active only during training (``mode=brainstate.mixin.Training``). In other
-    circumstances it is a no-op.
-
-    .. [1] Srivastava, Nitish, et al. "Dropout: a simple way to prevent
-           neural networks from overfitting." The journal of machine learning
-           research 15.1 (2014): 1929-1958.
-
-    .. admonition:: Tip
-        :class: tip
-
-        This kind of Dropout is firstly described in `Enabling Spike-based Backpropagation for Training Deep Neural
-        Network Architectures <https://arxiv.org/abs/1903.06379>`_:
-
-        There is a subtle difference in the way dropout is applied in SNNs compared to ANNs. In ANNs, each epoch of
-        training has several iterations of mini-batches. In each iteration, randomly selected units (with dropout ratio of :math:`p`)
-        are disconnected from the network while weighting by its posterior probability (:math:`1-p`). However, in SNNs, each
-        iteration has more than one forward propagation depending on the time length of the spike train. We back-propagate
-        the output error and modify the network parameters only at the last time step. For dropout to be effective in
-        our training method, it has to be ensured that the set of connected units within an iteration of mini-batch
-        data is not changed, such that the neural network is constituted by the same random subset of units during
-        each forward propagation within a single iteration. On the other hand, if the units are randomly connected at
-        each time-step, the effect of dropout will be averaged out over the entire forward propagation time within an
-        iteration. Then, the dropout effect would fade-out once the output error is propagated backward and the parameters
-        are updated at the last time step. Therefore, we need to keep the set of randomly connected units for the entire
-        time window within an iteration.
-
-    Args:
-      in_size: The size of the input tensor.
-      prob: Probability to keep element of the tensor.
-      mode: Mode. The computation mode of the object.
-      name: str. The name of the dynamic system.
-    """
-    __module__ = 'brainstate.nn'
-
-    def __init__(
-        self,
-        in_size: Size,
-        prob: float = 0.5,
-        name: Optional[str] = None
-    ) -> None:
-        super().__init__(name=name)
-        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
-        self.prob = prob
-        self.in_size = in_size
-        self.out_size = in_size
-
-    def init_state(self, batch_size=None, **kwargs):
-        if self.prob < 1.:
-            self.mask = ShortTermState(init.param(partial(random.bernoulli, self.prob), self.in_size, batch_size))
-
-    def update(self, x):
-        dtype = u.math.get_dtype(x)
-        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
-        if fit_phase and self.prob < 1.:
-            if self.mask.value.shape != x.shape:
-                raise ValueError(f"Input shape {x.shape} does not match the mask shape {self.mask.value.shape}. "
-                                 f"Please call `init_state()` method first.")
-            return u.math.where(self.mask.value,
-                                u.math.asarray(x / self.prob, dtype=dtype),
-                                u.math.asarray(0., dtype=dtype) * u.get_unit(x))
-        else:
-            return x
+# Copyright 2024 BrainX Ecosystem Limited. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+
+from functools import partial
+from typing import Optional, Sequence
+
+import brainunit as u
+import jax.numpy as jnp
+
+from brainstate import random, environ
+from brainstate._state import ShortTermState
+from brainstate.typing import Size
+from . import init as init
+from ._module import ElementWiseBlock
+
+__all__ = [
+    'Dropout',
+    'Dropout1d',
+    'Dropout2d',
+    'Dropout3d',
+    'AlphaDropout',
+    'FeatureAlphaDropout',
+    'DropoutFixed',
+]
+
+
+class Dropout(ElementWiseBlock):
+    """A layer that stochastically ignores a subset of inputs each training step.
+
+    In training, to compensate for the fraction of input values dropped (`rate`),
+    all surviving values are multiplied by `1 / (1 - rate)`.
+
+    This layer is active only during training (``mode=brainstate.mixin.Training``). In other
+    circumstances it is a no-op.
+
+    Parameters
+    ----------
+    prob : float
+        Probability to keep element of the tensor. Default is 0.5.
+    broadcast_dims : Sequence[int]
+        Dimensions that will share the same dropout mask. Default is ().
+    name : str, optional
+        The name of the dynamic system.
+
+    References
+    ----------
+    .. [1] Srivastava, Nitish, et al. "Dropout: a simple way to prevent
+           neural networks from overfitting." The journal of machine learning
+           research 15.1 (2014): 1929-1958.
+
+    Examples
+    --------
+    .. code-block:: python
+
+        >>> import brainstate
+        >>> layer = brainstate.nn.Dropout(prob=0.8)
+        >>> x = brainstate.random.randn(10, 20)
+        >>> with brainstate.environ.context(fit=True):
+        ...     output = layer(x)
+        >>> output.shape
+        (10, 20)
+
+    """
+    __module__ = 'brainstate.nn'
+
+    def __init__(
+        self,
+        prob: float = 0.5,
+        broadcast_dims: Sequence[int] = (),
+        name: Optional[str] = None
+    ) -> None:
+        super().__init__(name=name)
+        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
+        self.prob = prob
+        self.broadcast_dims = broadcast_dims
+
+    def __call__(self, x):
+        dtype = u.math.get_dtype(x)
+        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
+        if fit_phase and self.prob < 1.:
+            broadcast_shape = list(x.shape)
+            for dim in self.broadcast_dims:
+                broadcast_shape[dim] = 1
+            keep_mask = random.bernoulli(self.prob, broadcast_shape)
+            keep_mask = u.math.broadcast_to(keep_mask, x.shape)
+            return u.math.where(
+                keep_mask,
+                u.math.asarray(x / self.prob, dtype=dtype),
+                u.math.asarray(0., dtype=dtype)
+            )
+        else:
+            return x
+
+
+class _DropoutNd(ElementWiseBlock):
+    __module__ = 'brainstate.nn'
+    prob: float
+    channel_axis: int
+    minimal_dim: int
+
+    def __init__(
+        self,
+        prob: float = 0.5,
+        channel_axis: int = -1,
+        name: Optional[str] = None
+    ) -> None:
+        super().__init__(name=name)
+        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
+        self.prob = prob
+        self.channel_axis = channel_axis
+
+    def __call__(self, x):
+        # check input shape
+        inp_dim = u.math.ndim(x)
+        if inp_dim not in (self.minimal_dim, self.minimal_dim + 1):
+            raise RuntimeError(f"dropout1d: Expected {self.minimal_dim}D or {self.minimal_dim + 1}D input, "
+                               f"but received a {inp_dim}D input. {self._get_msg(x)}")
+        is_not_batched = self.minimal_dim
+        if is_not_batched:
+            channel_axis = self.channel_axis if self.channel_axis >= 0 else (x.ndim + self.channel_axis)
+            mask_shape = [(dim if i == channel_axis else 1) for i, dim in enumerate(x.shape)]
+        else:
+            channel_axis = (self.channel_axis + 1) if self.channel_axis >= 0 else (x.ndim + self.channel_axis)
+            assert channel_axis != 0, f"Channel axis must not be 0. But got {self.channel_axis}."
+            mask_shape = [(dim if i in (channel_axis, 0) else 1) for i, dim in enumerate(x.shape)]
+
+        # get fit phase
+        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
+
+        # generate mask
+        if fit_phase and self.prob < 1.:
+            dtype = u.math.get_dtype(x)
+            keep_mask = random.bernoulli(self.prob, mask_shape)
+            keep_mask = jnp.broadcast_to(keep_mask, x.shape)
+            return jnp.where(
+                keep_mask,
+                jnp.asarray(x / self.prob, dtype=dtype),
+                jnp.asarray(0., dtype=dtype)
+            )
+        else:
+            return x
+
+    def _get_msg(self, x):
+        return ''
+
+
+class Dropout1d(_DropoutNd):
+    r"""Randomly zero out entire channels (a channel is a 1D feature map).
+
+    Each channel will be zeroed out independently on every forward call with
+    probability using samples from a Bernoulli distribution. The channel is
+    a 1D feature map, e.g., the :math:`j`-th channel of the :math:`i`-th sample
+    in the batched input is a 1D tensor :math:`\text{input}[i, j]`.
+
+    Usually the input comes from :class:`Conv1d` modules.
+
+    As described in the paper [1]_, if adjacent pixels within feature maps are
+    strongly correlated (as is normally the case in early convolution layers)
+    then i.i.d. dropout will not regularize the activations and will otherwise
+    just result in an effective learning rate decrease.
+
+    In this case, :class:`Dropout1d` will help promote independence between
+    feature maps and should be used instead.
+
+    Parameters
+    ----------
+    prob : float
+        Probability of an element to be kept. Default is 0.5.
+    channel_axis : int
+        The axis representing the channel dimension. Default is -1.
+    name : str, optional
+        The name of the dynamic system.
+
+    Notes
+    -----
+    Input shape: :math:`(N, C, L)` or :math:`(C, L)`.
+
+    Output shape: :math:`(N, C, L)` or :math:`(C, L)` (same shape as input).
+
+    References
+    ----------
+    .. [1] Springenberg et al., "Striving for Simplicity: The All Convolutional Net"
+           https://arxiv.org/abs/1411.4280
+
+    Examples
+    --------
+    .. code-block:: python
+
+        >>> import brainstate
+        >>> m = brainstate.nn.Dropout1d(prob=0.8)
+        >>> x = brainstate.random.randn(20, 32, 16)
+        >>> with brainstate.environ.context(fit=True):
+        ...     output = m(x)
+        >>> output.shape
+        (20, 32, 16)
+
+    """
+    __module__ = 'brainstate.nn'
+    minimal_dim: int = 2
+
+    def _get_msg(self, x):
+        return ("Note that dropout1d exists to provide channel-wise dropout on inputs with 1 "
+                "spatial dimension, a channel dimension, and an optional batch dimension "
+                "(i.e. 2D or 3D inputs).")
+
+
+class Dropout2d(_DropoutNd):
+    r"""Randomly zero out entire channels (a channel is a 2D feature map).
+
+    Each channel will be zeroed out independently on every forward call with
+    probability using samples from a Bernoulli distribution. The channel is
+    a 2D feature map, e.g., the :math:`j`-th channel of the :math:`i`-th sample
+    in the batched input is a 2D tensor :math:`\text{input}[i, j]`.
+
+    Usually the input comes from :class:`Conv2d` modules.
+
+    As described in the paper [1]_, if adjacent pixels within feature maps are
+    strongly correlated (as is normally the case in early convolution layers)
+    then i.i.d. dropout will not regularize the activations and will otherwise
+    just result in an effective learning rate decrease.
+
+    In this case, :class:`Dropout2d` will help promote independence between
+    feature maps and should be used instead.
+
+    Parameters
+    ----------
+    prob : float
+        Probability of an element to be kept. Default is 0.5.
+    channel_axis : int
+        The axis representing the channel dimension. Default is -1.
+    name : str, optional
+        The name of the dynamic system.
+
+    Notes
+    -----
+    Input shape: :math:`(N, C, H, W)` or :math:`(C, H, W)`.
+
+    Output shape: :math:`(N, C, H, W)` or :math:`(C, H, W)` (same shape as input).
+
+    References
+    ----------
+    .. [1] Springenberg et al., "Striving for Simplicity: The All Convolutional Net"
+           https://arxiv.org/abs/1411.4280
+
+    Examples
+    --------
+    .. code-block:: python
+
+        >>> import brainstate
+        >>> m = brainstate.nn.Dropout2d(prob=0.8)
+        >>> x = brainstate.random.randn(20, 32, 32, 16)
+        >>> with brainstate.environ.context(fit=True):
+        ...     output = m(x)
+        >>> output.shape
+        (20, 32, 32, 16)
+
+    """
+    __module__ = 'brainstate.nn'
+    minimal_dim: int = 3
+
+    def _get_msg(self, x):
+        return ("Note that dropout2d exists to provide channel-wise dropout on inputs with 2 "
+                "spatial dimensions, a channel dimension, and an optional batch dimension "
+                "(i.e. 3D or 4D inputs).")
+
+
+class Dropout3d(_DropoutNd):
+    r"""Randomly zero out entire channels (a channel is a 3D feature map).
+
+    Each channel will be zeroed out independently on every forward call with
+    probability using samples from a Bernoulli distribution. The channel is
+    a 3D feature map, e.g., the :math:`j`-th channel of the :math:`i`-th sample
+    in the batched input is a 3D tensor :math:`\text{input}[i, j]`.
+
+    Usually the input comes from :class:`Conv3d` modules.
+
+    As described in the paper [1]_, if adjacent pixels within feature maps are
+    strongly correlated (as is normally the case in early convolution layers)
+    then i.i.d. dropout will not regularize the activations and will otherwise
+    just result in an effective learning rate decrease.
+
+    In this case, :class:`Dropout3d` will help promote independence between
+    feature maps and should be used instead.
+
+    Parameters
+    ----------
+    prob : float
+        Probability of an element to be kept. Default is 0.5.
+    channel_axis : int
+        The axis representing the channel dimension. Default is -1.
+    name : str, optional
+        The name of the dynamic system.
+
+    Notes
+    -----
+    Input shape: :math:`(N, C, D, H, W)` or :math:`(C, D, H, W)`.
+
+    Output shape: :math:`(N, C, D, H, W)` or :math:`(C, D, H, W)` (same shape as input).
+
+    References
+    ----------
+    .. [1] Springenberg et al., "Striving for Simplicity: The All Convolutional Net"
+           https://arxiv.org/abs/1411.4280
+
+    Examples
+    --------
+    .. code-block:: python
+
+        >>> import brainstate
+        >>> m = brainstate.nn.Dropout3d(prob=0.8)
+        >>> x = brainstate.random.randn(20, 16, 4, 32, 32)
+        >>> with brainstate.environ.context(fit=True):
+        ...     output = m(x)
+        >>> output.shape
+        (20, 16, 4, 32, 32)
+
+    """
+    __module__ = 'brainstate.nn'
+    minimal_dim: int = 4
+
+    def _get_msg(self, x):
+        return ("Note that dropout3d exists to provide channel-wise dropout on inputs with 3 "
+                "spatial dimensions, a channel dimension, and an optional batch dimension "
+                "(i.e. 4D or 5D inputs).")
+
+
+class AlphaDropout(_DropoutNd):
+    r"""Applies Alpha Dropout over the input.
+
+    Alpha Dropout is a type of Dropout that maintains the self-normalizing
+    property. For an input with zero mean and unit standard deviation, the output of
+    Alpha Dropout maintains the original mean and standard deviation of the
+    input.
+
+    Alpha Dropout goes hand-in-hand with SELU activation function, which ensures
+    that the outputs have zero mean and unit standard deviation.
+
+    During training, it randomly masks some of the elements of the input
+    tensor with probability using samples from a Bernoulli distribution.
+    The elements to be masked are randomized on every forward call, and scaled
+    and shifted to maintain zero mean and unit standard deviation.
+
+    During evaluation the module simply computes an identity function.
+
+    Parameters
+    ----------
+    prob : float
+        Probability of an element to be kept. Default is 0.5.
+    name : str, optional
+        The name of the dynamic system.
+
+    Notes
+    -----
+    Input shape: :math:`(*)`. Input can be of any shape.
+
+    Output shape: :math:`(*)`. Output is of the same shape as input.
+
+    References
+    ----------
+    .. [1] Klambauer et al., "Self-Normalizing Neural Networks"
+           https://arxiv.org/abs/1706.02515
+
+    Examples
+    --------
+    .. code-block:: python
+
+        >>> import brainstate
+        >>> m = brainstate.nn.AlphaDropout(prob=0.8)
+        >>> x = brainstate.random.randn(20, 16)
+        >>> with brainstate.environ.context(fit=True):
+        ...     output = m(x)
+        >>> output.shape
+        (20, 16)
+
+    """
+    __module__ = 'brainstate.nn'
+
+    def __init__(
+        self,
+        prob: float = 0.5,
+        name: Optional[str] = None
+    ) -> None:
+        super().__init__(name=name)
+        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
+        self.prob = prob
+
+        # SELU parameters
+        alpha = -1.7580993408473766
+        self.alpha = alpha
+
+        # Affine transformation parameters to maintain mean and variance
+        self.a = ((1 - prob) * (1 + prob * alpha ** 2)) ** -0.5
+        self.b = -self.a * alpha * prob
+
+    def __call__(self, x):
+        dtype = u.math.get_dtype(x)
+        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
+        if fit_phase and self.prob < 1.:
+            keep_mask = random.bernoulli(self.prob, x.shape)
+            return u.math.where(
+                keep_mask,
+                u.math.asarray(x, dtype=dtype),
+                u.math.asarray(self.alpha, dtype=dtype)
+            ) * self.a + self.b
+        else:
+            return x
+
+
+class FeatureAlphaDropout(ElementWiseBlock):
+    r"""Randomly masks out entire channels with Alpha Dropout properties.
+
+    Instead of setting activations to zero as in regular Dropout, the activations
+    are set to the negative saturation value of the SELU activation function to
+    maintain self-normalizing properties.
+
+    Each channel (e.g., the :math:`j`-th channel of the :math:`i`-th sample in
+    the batch input is a tensor :math:`\text{input}[i, j]`) will be masked
+    independently for each sample on every forward call with probability using
+    samples from a Bernoulli distribution. The elements to be masked are randomized
+    on every forward call, and scaled and shifted to maintain zero mean and unit
+    variance.
+
+    Usually the input comes from convolutional layers with SELU activation.
+
+    As described in the paper [2]_, if adjacent pixels within feature maps are
+    strongly correlated (as is normally the case in early convolution layers)
+    then i.i.d. dropout will not regularize the activations and will otherwise
+    just result in an effective learning rate decrease.
+
+    In this case, :class:`FeatureAlphaDropout` will help promote independence between
+    feature maps and should be used instead.
+
+    Parameters
+    ----------
+    prob : float
+        Probability of an element to be kept. Default is 0.5.
+    channel_axis : int
+        The axis representing the channel dimension. Default is -1.
+    name : str, optional
+        The name of the dynamic system.
+
+    Notes
+    -----
+    Input shape: :math:`(N, C, *)` where C is the channel dimension.
+
+    Output shape: Same shape as input.
+
+    References
+    ----------
+    .. [1] Klambauer et al., "Self-Normalizing Neural Networks"
+           https://arxiv.org/abs/1706.02515
+    .. [2] Springenberg et al., "Striving for Simplicity: The All Convolutional Net"
+           https://arxiv.org/abs/1411.4280
+
+    Examples
+    --------
+    .. code-block:: python
+
+        >>> import brainstate
+        >>> m = brainstate.nn.FeatureAlphaDropout(prob=0.8)
+        >>> x = brainstate.random.randn(20, 16, 4, 32, 32)
+        >>> with brainstate.environ.context(fit=True):
+        ...     output = m(x)
+        >>> output.shape
+        (20, 16, 4, 32, 32)
+
+    """
+    __module__ = 'brainstate.nn'
+
+    def __init__(
+        self,
+        prob: float = 0.5,
+        channel_axis: int = -1,
+        name: Optional[str] = None
+    ) -> None:
+        super().__init__(name=name)
+        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
+        self.prob = prob
+        self.channel_axis = channel_axis
+
+        # SELU parameters
+        alpha = -1.7580993408473766
+        self.alpha = alpha
+
+        # Affine transformation parameters to maintain mean and variance
+        self.a = ((1 - prob) * (1 + prob * alpha ** 2)) ** -0.5
+        self.b = -self.a * alpha * prob
+
+    def __call__(self, x):
+        dtype = u.math.get_dtype(x)
+        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
+        if fit_phase and self.prob < 1.:
+            # Create mask shape with 1s except for batch and channel dimensions
+            channel_axis = self.channel_axis if self.channel_axis >= 0 else (x.ndim + self.channel_axis)
+            mask_shape = [1] * x.ndim
+            mask_shape[0] = x.shape[0]  # batch dimension
+            mask_shape[channel_axis] = x.shape[channel_axis]  # channel dimension
+
+            keep_mask = random.bernoulli(self.prob, mask_shape)
+            keep_mask = u.math.broadcast_to(keep_mask, x.shape)
+            return u.math.where(
+                keep_mask,
+                u.math.asarray(x, dtype=dtype),
+                u.math.asarray(self.alpha, dtype=dtype)
+            ) * self.a + self.b
+        else:
+            return x
+
+
+class DropoutFixed(ElementWiseBlock):
+    """A dropout layer with a fixed dropout mask along the time axis.
+
+    In training, to compensate for the fraction of input values dropped,
+    all surviving values are multiplied by `1 / (1 - prob)`.
+
+    This layer is active only during training (``mode=brainstate.mixin.Training``). In other
+    circumstances it is a no-op.
+
+    This kind of Dropout is particularly useful for spiking neural networks (SNNs) where
+    the same dropout mask needs to be applied across multiple time steps within a single
+    mini-batch iteration.
+
+    Parameters
+    ----------
+    in_size : tuple or int
+        The size of the input tensor.
+    prob : float
+        Probability to keep element of the tensor. Default is 0.5.
+    name : str, optional
+        The name of the dynamic system.
+
+    Notes
+    -----
+    As described in [2]_, there is a subtle difference in the way dropout is applied in
+    SNNs compared to ANNs. In ANNs, each epoch of training has several iterations of
+    mini-batches. In each iteration, randomly selected units (with dropout ratio of
+    :math:`p`) are disconnected from the network while weighting by its posterior
+    probability (:math:`1-p`).
+
+    However, in SNNs, each iteration has more than one forward propagation depending on
+    the time length of the spike train. We back-propagate the output error and modify
+    the network parameters only at the last time step. For dropout to be effective in
+    our training method, it has to be ensured that the set of connected units within an
+    iteration of mini-batch data is not changed, such that the neural network is
+    constituted by the same random subset of units during each forward propagation within
+    a single iteration.
+
+    On the other hand, if the units are randomly connected at each time-step, the effect
+    of dropout will be averaged out over the entire forward propagation time within an
+    iteration. Then, the dropout effect would fade-out once the output error is propagated
+    backward and the parameters are updated at the last time step. Therefore, we need to
+    keep the set of randomly connected units for the entire time window within an iteration.
+
+    References
+    ----------
+    .. [1] Srivastava, Nitish, et al. "Dropout: a simple way to prevent
+           neural networks from overfitting." The journal of machine learning
+           research 15.1 (2014): 1929-1958.
+    .. [2] Lee et al., "Enabling Spike-based Backpropagation for Training Deep Neural
+           Network Architectures" https://arxiv.org/abs/1903.06379
+
+    Examples
+    --------
+    .. code-block:: python
+
+        >>> import brainstate
+        >>> layer = brainstate.nn.DropoutFixed(in_size=(20,), prob=0.8)
+        >>> layer.init_state(batch_size=10)
+        >>> x = brainstate.random.randn(10, 20)
+        >>> with brainstate.environ.context(fit=True):
+        ...     output = layer.update(x)
+        >>> output.shape
+        (10, 20)
+
+    """
+    __module__ = 'brainstate.nn'
+
+    def __init__(
+        self,
+        in_size: Size,
+        prob: float = 0.5,
+        name: Optional[str] = None
+    ) -> None:
+        super().__init__(name=name)
+        assert 0. <= prob <= 1., f"Dropout probability must be in the range [0, 1]. But got {prob}."
+        self.prob = prob
+        self.in_size = in_size
+        self.out_size = in_size
+
+    def init_state(self, batch_size=None, **kwargs):
+        if self.prob < 1.:
+            self.mask = ShortTermState(init.param(partial(random.bernoulli, self.prob), self.in_size, batch_size))
+
+    def update(self, x):
+        dtype = u.math.get_dtype(x)
+        fit_phase = environ.get('fit', desc='Whether this is a fitting process. Bool.')
+        if fit_phase and self.prob < 1.:
+            if self.mask.value.shape != x.shape:
+                raise ValueError(f"Input shape {x.shape} does not match the mask shape {self.mask.value.shape}. "
+                                 f"Please call `init_state()` method first.")
+            return u.math.where(self.mask.value,
+                                u.math.asarray(x / self.prob, dtype=dtype),
+                                u.math.asarray(0., dtype=dtype) * u.get_unit(x))
+        else:
+            return x